Copilot Memory Prompt: Save the Context You Need Next

A Copilot memory prompt is a structured instruction for reviewing a conversation and extracting the information worth keeping before the session ends. Instead of trying to preserve an entire chat, it identifies the specific facts confirmed, the decisions made, and the next steps waiting — the things that actually matter when you or a colleague returns to the work. Tools like fenbs provide a place to store that extracted context outside Copilot so the next session — in any tool — starts from a useful record rather than a blank slate.

This page focuses on the saving prompt itself: how to write it, when to run it, and what to do with the output. For detail on Copilot's own memory features — what it retains across sessions and how that compares to a manual approach — see Microsoft Copilot memory and does Copilot have memory.

Why saving a summary beats keeping the full chat

A full conversation contains the same information several times over: the initial question, the AI's first attempt, your correction, the revised answer, another refinement. What matters to the next session is not the negotiation — it is the settled outcome. Saving the full transcript preserves that process at the cost of making the settled outcome harder to find.

A well-structured summary saves only three things: facts that were confirmed during the session, decisions that were made and the reason behind them, and the next steps that follow. This is compact enough to paste at the start of a new session, clear enough for a teammate to read, and precise enough for a different AI tool to use without misreading the intent.

The same principle applies to any AI assistant, not just Copilot. If you use Claude, ChatGPT, or Gemini for similar work, the saving prompt below will work there too. The challenge is always the same: a useful record is smaller than the conversation that produced it.

The Copilot memory prompt

Run this prompt near the end of a Copilot session, after the main work is done. It produces a structured output you can review and then save or paste wherever you keep work context.

Before I close this session, help me save what matters for next time.

Review our conversation and produce a handoff summary in this format:

PROJECT / TASK: [name of the project or task we worked on]

FACTS CONFIRMED:
— [fact that was established or verified during this session]
— [repeat for each key fact]

DECISIONS MADE:
— [decision]: [what was decided and why, in one sentence]
— [repeat for each decision]

OPEN QUESTIONS:
— [question that came up but was not resolved]

NEXT STEPS:
— [specific action]: [who needs to do it / what it requires]
— [repeat for each next step]

WHERE WE LEFT OFF:
[One paragraph: what was completed, what remains, and where the next session should start]

Do not include: opinions, alternative options that were considered but not chosen, or general background information already in the project notes.

The instruction at the end to exclude background information is important. A summary that repeats what you already know is longer without being more useful. Only information that is new, settled, or time-sensitive belongs in the handoff.

A shorter version for quick sessions

For a short Copilot session — twenty minutes of drafting or a quick research question — the full handoff prompt is more than you need. Use this shorter version when the session was focused and the context is already well established.

Summarize what we settled in this session in three bullet points:
1. What we decided or confirmed
2. What still needs to happen
3. Where the next session should start

Keep each point to one sentence.

Paste the output of either prompt into your project notes at the end of the session. When you return, read the summary before starting rather than asking Copilot to remind you of where you were — the summary is more reliable than the model's own account of a previous session.

What to do with the summary

The summary has three likely uses. First, you paste it at the start of the next session so the tool reads the current state before proceeding. Second, you share it with a teammate so they can continue without asking you to debrief. Third, you add the decisions to a longer-running project record so they are not lost when the session closes.

For each of these, the value of the summary depends on its accuracy. Read it before saving. A summary that misrepresents a decision — for example, treating "we should consider this" as "we decided this" — is worse than no summary because the next session will act on a false record. Edit anything that is not accurate before you save it.

How to use the summary in the next session

The value of saving context is realized when the next session reads it correctly. Paste the summary at the start of the next session before any task, along with a brief instruction to use it.

Here is where we left off on [project name]:

[paste the saved summary]

Before we continue, confirm you understand:
1. What was decided in the last session
2. What is still open
3. Where you will start

Then tell me what you need from me to take the next step.

This confirmation step at the start is worth running at least the first few times. A model that misreads the summary — treating an open question as a settled decision, or starting at the wrong step — will do so silently unless you check. After a few sessions you will have a sense of how reliably the tool reads your summaries and can decide how much checking is still needed.

How to do this with fenbs

1. Create a free account and add a project for the work you are doing in Copilot.
2. Save the full session-close prompt and the quick summary prompt as Skills in the project.
3. At the end of each Copilot session, run the appropriate saving prompt and paste the output into the project's Where we left off section or as a context note.
4. Connect a supported AI app over MCP from Integrations. At the start of the next session, ask the assistant to read the project context — it will see the summary without you pasting it.
5. When a decision is confirmed in a Copilot session, add it to fenbs as a Decision with the date and reason, so it is not lost if you never return to that specific session.

fenbs does not read your Copilot history or connect to Microsoft 365 directly. You run the saving prompt in Copilot, review the output, and paste or record it in fenbs. The separation is deliberate: fenbs keeps the settled conclusions; the full conversation stays in Copilot if you need to refer back to it.